The Research On Prediction Of Water Inrush From Coal Seam Floor Under Unbalanced Data

Mine water inrush is one of the major disasters in coal mine production,and its sudden and destructive power is characterized by frequent occurrence of accidents,which threaten the safety of the miners' life and property.Due to the influence of natural environment hydrogeology and other complicated factors,the prediction effect of traditional method is not ideal.Some emerging intelligent algorithms make water inrush prediction under the balance sample.Although some effects have been achieved,there are still some problems in the water inrush prediction under balanced samples.In reality,the water inrush event is far less than that of non-water inrush,and the number of water inrush samples is very rare,so that the sample set is often in an unbalanced state.If the prediction accuracy of water inrush event is higher,the prediction accuracy of water inrush event is lower,which will directly affect the prediction effect.Therefore,the prediction precision of a few samples under unbalanced data is a problem that needs to be solved in the field of water inrush prediction of coal seam floor.Based on the sufficient research on the basis of traditional method and machine learning algorithm,In view of the characteristics of water inrush data of coal seam floor,a k-mean-relisthsmote-pso-svm based prediction model of water inrush is constructed under the unbalanced data Firstly,the K-means method was applied to optimize the Relief method,and the water inrush index was screened to make up the deficiency of the weight value of the water inrush feature due to the Relief index screening method.Secondly,in view of the SMOTE method in dealing with unbalanced data because of noise sensitive data interpolation fuzzy positive and negative boundary space is too small lead to problems such as fitting,h d of sampling algorithm(HSMOTE),make the inrush data set into balance.Application particle swarm optimization support vector machine(pso-svm)is used to predict water in burst data.In the Matlab platform,the fertility German data set in UCI database was applied for simulation experiment to verify the feasibility and accuracy of K-means-Relief-HSMOTE-PSO-SVM model.Through the analysis of the samples of typical coal mine in north China,the main influencing factors of water inrush of six coal seam floor are selected.It was tested 50 times in k-means-Relief-HSMOTE-PSO-SVM and compared with other models.The results show that the prediction model of water inrush of coal seam floor is better than that of comparison model.The model is able to extract the key inwater feature factor and make up the defect of SMOTE method,and optimize the parameter selection of SVM.The prediction accuracy and geometric average accuracy of a few types of water inrush samples are effectively improved.